tcm_vhost: Multi-queue support
Subsystems (3)
PATCH 1/1 tcm_vhost: Multi-queue support
This adds virtio-scsi multi-queue support to tcm_vhost.
Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166
Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)
drivers/vhost/tcm_vhost.c +29 -17
@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h"
@@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */
@@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */
@@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,
@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(
@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;
@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );
@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {
@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/*
@@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)
@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;
@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)
drivers/vhost/tcm_vhost.h +2 -0
@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Signature · 1 line
1.8.1
Quoted 4 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
Quoted 9 lines
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
Nice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN tcm_vhost endpoints as well.. 8-)
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Quoted 4 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
Quoted 9 lines
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
Nice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN tcm_vhost endpoints as well.. 8-)
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Quoted 11 lines · Nicholas A. Bellinger
Quoted 10 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS
Quoted 8 lines
Quoted 3 lines
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPSNice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN
tcm_vhostendpoints as well.. 8-)
Do you mean something like this?
1 queue, 2 targets, 2 lun per target
4 queue, 2 targets, 2 lun per target
Quoted 3 lines
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
Ah, this applies on top of mst's 'tcm_vhost: fix pr_err on early kick patch.' plus my last v3 of 'tcm_vhost: Multi-target support'.
Quoted 168 lines
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Signature · 1 line
Asias
Quoted 26 lines · Asias He
Quoted 11 lines · Nicholas A. Bellinger
Quoted 10 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPSQuoted 8 lines
Quoted 3 lines
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPSNice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN
tcm_vhostendpoints as well.. 8-)Do you mean something like this?
1 queue, 2 targets, 2 lun per target
4 queue, 2 targets, 2 lun per target
Quoted 7 lines
Quoted 3 lines
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
Ah, this applies on top of mst's '
tcm_vhost: fixpr_erron early kick patch.' plus my last v3 of 'tcm_vhost: Multi-target support'.
In that case, applying this patch + PATCH-v3 to auto-next for testing for the moment, and will respin for-next against upstream w/ MST's patch shortly.
Also, please include a proper changelog for this second patch. :)
Thank you!
--nab
Quoted 170 lines
Quoted 168 lines
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Quoted 31 lines · Nicholas A. Bellinger
Quoted 30 lines · Asias He
Quoted 11 lines · Nicholas A. Bellinger
Quoted 10 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPSQuoted 8 lines
Quoted 3 lines
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPSNice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN
tcm_vhostendpoints as well.. 8-)Do you mean something like this?
1 queue, 2 targets, 2 lun per target
4 queue, 2 targets, 2 lun per targetQuoted 3 lines
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
Quoted 7 lines
Quoted 3 lines
Ah, this applies on top of mst's '
tcm_vhost: fixpr_erron early kick patch.' plus my last v3 of 'tcm_vhost: Multi-target support'.In that case, applying this patch + PATCH-v3 to auto-next for testing for the moment, and will respin for-next against upstream w/ MST's patch shortly.
Okay. Looking forward to more perf numbers.
Quoted 1 line
Also, please include a proper changelog for this second patch. :)
Sure.
---->
tcm_vhost: Multi-queue support
This adds virtio-scsi multi-queue support to tcm_vhost. In order to use multi-queue, guest side multi-queue support is need. It can be found here:
https://lkml.org/lkml/2012/12/18/166
Currently, only one thread is created by vhost core code for each vhost_scsi instance. Even if there are multi-queues, all the handling of guest kick (vhost_scsi_handle_kick) are processed in one thread. This is not optimal. Luckily, most of the work is offloaded to the tcm_vhost workqueue.
Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
<----
Quoted 178 lines
Thank you!
--nab
Quoted 170 lines
Quoted 168 lines
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */
Signature · 1 line
Asias
Paolo Bonzini
maintainer----- Messaggio originale -----
Quoted 8 lines
Da: "Asias He" <[email protected]>
A: "Nicholas A. Bellinger" <[email protected]>
Cc: "Paolo Bonzini" <[email protected]>, "Stefan Hajnoczi" <[email protected]>, "Michael S. Tsirkin" <[email protected]>, "Rusty Russell" <[email protected]>, [email protected], [email protected], [email protected]
Inviato: Mercoledì, 6 febbraio 2013 10:51:34
Oggetto: Re: [PATCH]tcm_vhost: Multi-queue support
Quoted 16 lines
Quoted 15 lines · Nicholas A. Bellinger
Quoted 14 lines · Asias He
Quoted 13 lines · Nicholas A. Bellinger
Quoted 12 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
4 VCPUs I suppose?
Paolo
Quoted 21 lines · Paolo Bonzini
----- Messaggio originale -----
Quoted 18 lines
Da: "Asias He" <[email protected]>
A: "Nicholas A. Bellinger" <[email protected]>
Cc: "Paolo Bonzini" <[email protected]>, "Stefan Hajnoczi" <[email protected]>, "Michael S. Tsirkin" <[email protected]>, "Rusty Russell" <[email protected]>, [email protected], [email protected], [email protected]
Inviato: Mercoledì, 6 febbraio 2013 10:51:34
Oggetto: Re: [PATCH]tcm_vhost: Multi-queue supportQuoted 9 lines · Nicholas A. Bellinger
Quoted 8 lines · Asias He
Quoted 7 lines · Nicholas A. Bellinger
Quoted 6 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS4 VCPUs I suppose?
Yes.
Signature · 5 lines
Asias _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Paolo Bonzini
maintainer----- Messaggio originale -----
Quoted 8 lines
Da: "Asias He" <[email protected]>
A: "Nicholas A. Bellinger" <[email protected]>
Cc: "Paolo Bonzini" <[email protected]>, "Stefan Hajnoczi" <[email protected]>, "Michael S. Tsirkin" <[email protected]>, "Rusty Russell" <[email protected]>, [email protected], [email protected], [email protected]
Inviato: Mercoledì, 6 febbraio 2013 10:51:34
Oggetto: Re: [PATCH]tcm_vhost: Multi-queue support
Quoted 16 lines
Quoted 15 lines · Nicholas A. Bellinger
Quoted 14 lines · Asias He
Quoted 13 lines · Nicholas A. Bellinger
Quoted 12 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPS4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPS
4 VCPUs I suppose?
Paolo
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Quoted 26 lines · Asias He
Quoted 11 lines · Nicholas A. Bellinger
Quoted 10 lines · Asias He
This adds virtio-scsi multi-queue support to
tcm_vhost.Guest side virtio-scsi multi-queue support can be found here:
https://lkml.org/lkml/2012/12/18/166Some initial perf numbers:
1 queue, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 127K/127k IOPSQuoted 8 lines
Quoted 3 lines
4 queues, 4 targets, 1 lun per target
4K request size, 50% randread + 50% randwrite: 181K/181k IOPSNice single LUN small block random I/O improvement here with 4x vqueues.
Curious to see how virtio-scsi small block performance looks with SCSI-core to multi-LUN
tcm_vhostendpoints as well.. 8-)Do you mean something like this?
1 queue, 2 targets, 2 lun per target
4 queue, 2 targets, 2 lun per target
Quoted 7 lines
Quoted 3 lines
Btw, this does not apply atop current target-pending.git/for-next with your other pending vhost patch series, and AFAICT this patch is supposed to apply on top of your last PATCH-v3, no..?
Ah, this applies on top of mst's '
tcm_vhost: fixpr_erron early kick patch.' plus my last v3 of 'tcm_vhost: Multi-target support'.
In that case, applying this patch + PATCH-v3 to auto-next for testing for the moment, and will respin for-next against upstream w/ MST's patch shortly.
Also, please include a proper changelog for this second patch. :)
Thank you!
--nab
Quoted 170 lines
Quoted 168 lines
--nab
Quoted 164 lines
Signed-off-by Asias He <[email protected]>--- drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- drivers/vhost/tcm_vhost.h | 2 ++ 2 files changed, 31 insertions(+), 17 deletions(-)drivers/vhost/tcm_vhost.c +29
-17@@ -48,6 +48,7 @@#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_scsi.h> #include <linux/llist.h> +#include <linux/bitmap.h> #include "vhost.c" #include "vhost.h" @@ -59,7 +60,8 @@ enum { VHOST_SCSI_VQ_IO = 2, }; -#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_TARGET 256 +#define VHOST_SCSI_MAX_VQ 128struct vhost_scsi {/* Protected by vhost_scsi->dev.mutex */ @@ -68,7 +70,7 @@ struct vhost_scsi {bool vs_endpoint;struct vhost_dev dev; -struct vhost_virtqueue vqs[3 ]; +struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ ];struct vhost_work vs_completion_work;/* cmd completion work item */ struct llist_head vs_completion_list;/* cmd completion queue */ @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) { struct vhost_scsi *vs = container_of(work, struct vhost_scsi, vs_completion_work); +DECLARE_BITMAP (signal,VHOST_SCSI_MAX_VQ );struct virtio_scsi_cmd_resp v_rsp;struct tcm_vhost_cmd *tv_cmd;struct llist_node *llnode;struct se_cmd *se_cmd; -int ret; +int ret, vq; +bitmap_zero (signal,VHOST_SCSI_MAX_VQ ); llnode =llist_del_all (&vs->vs_completion_list);while (llnode) { tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)memcpy (v_rsp.sense, tv_cmd->tvc_sense_buf, v_rsp.sense_len); ret =copy_to_user (tv_cmd->tvc_resp, &v_rsp,sizeof (v_rsp)); -if (likely (ret ==0 )) -vhost_add_used (&vs->vqs[2 ], tv_cmd->tvc_vq_desc,0 ); -else +if (likely (ret ==0 )) { +vhost_add_used (tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc,0 ); + vq = tv_cmd->tvc_vq - vs->vqs; +__set_bit (vq, signal); + }else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );vhost_scsi_free_cmd (tv_cmd); } -vhost_signal (&vs->dev, &vs->vqs[2 ]); + vq =-1 ; +while ((vq =find_next_bit (signal,VHOST_SCSI_MAX_VQ , vq +1 )) + <VHOST_SCSI_MAX_VQ ) +vhost_signal (&vs->dev, &vs->vqs[vq]); } static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) } } -static void vhost_scsi_handle_vq (struct vhost_scsi *vs) +static void vhost_scsi_handle_vq (struct vhost_scsi *vs, +struct vhost_virtqueue *vq) { -struct vhost_virtqueue *vq = &vs->vqs[2 ];struct virtio_scsi_cmd_req v_req;struct tcm_vhost_tpg *tv_tpg;struct tcm_vhost_cmd *tv_cmd;@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ret =__copy_to_user (resp, &rsp,sizeof (rsp));if (!ret)vhost_add_used_and_signal (&vs->dev, - &vs->vqs[2 ], head,0 ); + vq, head,0 );else pr_err ("Faulted on virtio_scsi_cmd_resp\n" );@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ": %d\n", tv_cmd, exp_data_len, data_direction); tv_cmd->tvc_vhost = vs; + tv_cmd->tvc_vq = vq;if (unlikely (vq->iov[out].iov_len !=sizeof (struct virtio_scsi_cmd_resp ))) {@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) poll.work); struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); -vhost_scsi_handle_vq (vs); +vhost_scsi_handle_vq (vs, vq); }/* @@ -879,7 +889,7 @@ err:static int vhost_scsi_open (struct inode *inode,struct file *f) {struct vhost_scsi *s; -int r; +int r, i; s =kzalloc (sizeof (*s),GFP_KERNEL );if (!s)@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) s->vqs[VHOST_SCSI_VQ_CTL ].handle_kick = vhost_scsi_ctl_handle_kick; s->vqs[VHOST_SCSI_VQ_EVT ].handle_kick = vhost_scsi_evt_handle_kick; - s->vqs[VHOST_SCSI_VQ_IO ].handle_kick = vhost_scsi_handle_kick; - r =vhost_dev_init (&s->dev, s->vqs,3 ); +for (i =VHOST_SCSI_VQ_IO ; i <VHOST_SCSI_MAX_VQ ; i++) + s->vqs[i].handle_kick = vhost_scsi_handle_kick; + r =vhost_dev_init (&s->dev, s->vqs,VHOST_SCSI_MAX_VQ );if (r <0 ) {kfree (s);return r;@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)static void vhost_scsi_flush (struct vhost_scsi *vs) { -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_CTL ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_EVT ); -vhost_scsi_flush_vq (vs,VHOST_SCSI_VQ_IO ); +int i; + +for (i =0 ; i <VHOST_SCSI_MAX_VQ ; i++) +vhost_scsi_flush_vq (vs, i); }static int vhost_scsi_set_features (struct vhost_scsi *vs,u64 features)drivers/vhost/tcm_vhost.h +2
-0@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {struct virtio_scsi_cmd_resp __user *tvc_resp;/* Pointer to vhost_scsi for our device */ struct vhost_scsi *tvc_vhost; +/* Pointer to vhost_virtqueue for the cmd */ +struct vhost_virtqueue *tvc_vq;/* Pointer to vhost nexus memory */ struct tcm_vhost_nexus *tvc_nexus;/* The TCM I/O descriptor that is accessed via container_of() */